1#ifndef _LINUX_TRACEPOINT_H
2#define _LINUX_TRACEPOINT_H
3
4/*
5 * Kernel Tracepoint API.
6 *
7 * See Documentation/trace/tracepoints.txt.
8 *
9 * Copyright (C) 2008-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
10 *
11 * Heavily inspired from the Linux Kernel Markers.
12 *
13 * This file is released under the GPLv2.
14 * See the file COPYING for more details.
15 */
16
17#include <linux/smp.h>
18#include <linux/errno.h>
19#include <linux/types.h>
20#include <linux/cpumask.h>
21#include <linux/rcupdate.h>
22#include <linux/static_key.h>
23
24struct module;
25struct tracepoint;
26struct notifier_block;
27
28struct tracepoint_func {
29	void *func;
30	void *data;
31};
32
33struct tracepoint {
34	const char *name;		/* Tracepoint name */
35	struct static_key key;
36	void (*regfunc)(void);
37	void (*unregfunc)(void);
38	struct tracepoint_func __rcu *funcs;
39};
40
41struct trace_enum_map {
42	const char		*system;
43	const char		*enum_string;
44	unsigned long		enum_value;
45};
46
47extern int
48tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
49extern int
50tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
51extern void
52for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
53		void *priv);
54
55#ifdef CONFIG_MODULES
56struct tp_module {
57	struct list_head list;
58	struct module *mod;
59};
60
61bool trace_module_has_bad_taint(struct module *mod);
62extern int register_tracepoint_module_notifier(struct notifier_block *nb);
63extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
64#else
65static inline bool trace_module_has_bad_taint(struct module *mod)
66{
67	return false;
68}
69static inline
70int register_tracepoint_module_notifier(struct notifier_block *nb)
71{
72	return 0;
73}
74static inline
75int unregister_tracepoint_module_notifier(struct notifier_block *nb)
76{
77	return 0;
78}
79#endif /* CONFIG_MODULES */
80
81/*
82 * tracepoint_synchronize_unregister must be called between the last tracepoint
83 * probe unregistration and the end of module exit to make sure there is no
84 * caller executing a probe when it is freed.
85 */
86static inline void tracepoint_synchronize_unregister(void)
87{
88	synchronize_sched();
89}
90
91#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
92extern void syscall_regfunc(void);
93extern void syscall_unregfunc(void);
94#endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
95
96#define PARAMS(args...) args
97
98#define TRACE_DEFINE_ENUM(x)
99
100#endif /* _LINUX_TRACEPOINT_H */
101
102/*
103 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
104 *  file ifdef protection.
105 *  This is due to the way trace events work. If a file includes two
106 *  trace event headers under one "CREATE_TRACE_POINTS" the first include
107 *  will override the TRACE_EVENT and break the second include.
108 */
109
110#ifndef DECLARE_TRACE
111
112#define TP_PROTO(args...)	args
113#define TP_ARGS(args...)	args
114#define TP_CONDITION(args...)	args
115
116#ifdef CONFIG_TRACEPOINTS
117
118/*
119 * it_func[0] is never NULL because there is at least one element in the array
120 * when the array itself is non NULL.
121 *
122 * Note, the proto and args passed in includes "__data" as the first parameter.
123 * The reason for this is to handle the "void" prototype. If a tracepoint
124 * has a "void" prototype, then it is invalid to declare a function
125 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
126 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
127 */
128#define __DO_TRACE(tp, proto, args, cond, prercu, postrcu)		\
129	do {								\
130		struct tracepoint_func *it_func_ptr;			\
131		void *it_func;						\
132		void *__data;						\
133									\
134		if (!(cond))						\
135			return;						\
136		prercu;							\
137		rcu_read_lock_sched_notrace();				\
138		it_func_ptr = rcu_dereference_sched((tp)->funcs);	\
139		if (it_func_ptr) {					\
140			do {						\
141				it_func = (it_func_ptr)->func;		\
142				__data = (it_func_ptr)->data;		\
143				((void(*)(proto))(it_func))(args);	\
144			} while ((++it_func_ptr)->func);		\
145		}							\
146		rcu_read_unlock_sched_notrace();			\
147		postrcu;						\
148	} while (0)
149
150#ifndef MODULE
151#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)	\
152	static inline void trace_##name##_rcuidle(proto)		\
153	{								\
154		if (static_key_false(&__tracepoint_##name.key))		\
155			__DO_TRACE(&__tracepoint_##name,		\
156				TP_PROTO(data_proto),			\
157				TP_ARGS(data_args),			\
158				TP_CONDITION(cond),			\
159				rcu_irq_enter(),			\
160				rcu_irq_exit());			\
161	}
162#else
163#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
164#endif
165
166/*
167 * Make sure the alignment of the structure in the __tracepoints section will
168 * not add unwanted padding between the beginning of the section and the
169 * structure. Force alignment to the same alignment as the section start.
170 *
171 * When lockdep is enabled, we make sure to always do the RCU portions of
172 * the tracepoint code, regardless of whether tracing is on or we match the
173 * condition.  This lets us find RCU issues triggered with tracepoints even
174 * when this tracepoint is off.  This code has no purpose other than poking
175 * RCU a bit.
176 */
177#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
178	extern struct tracepoint __tracepoint_##name;			\
179	static inline void trace_##name(proto)				\
180	{								\
181		if (static_key_false(&__tracepoint_##name.key))		\
182			__DO_TRACE(&__tracepoint_##name,		\
183				TP_PROTO(data_proto),			\
184				TP_ARGS(data_args),			\
185				TP_CONDITION(cond),,);			\
186		if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {		\
187			rcu_read_lock_sched_notrace();			\
188			rcu_dereference_sched(__tracepoint_##name.funcs);\
189			rcu_read_unlock_sched_notrace();		\
190		}							\
191	}								\
192	__DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args),		\
193		PARAMS(cond), PARAMS(data_proto), PARAMS(data_args))	\
194	static inline int						\
195	register_trace_##name(void (*probe)(data_proto), void *data)	\
196	{								\
197		return tracepoint_probe_register(&__tracepoint_##name,	\
198						(void *)probe, data);	\
199	}								\
200	static inline int						\
201	unregister_trace_##name(void (*probe)(data_proto), void *data)	\
202	{								\
203		return tracepoint_probe_unregister(&__tracepoint_##name,\
204						(void *)probe, data);	\
205	}								\
206	static inline void						\
207	check_trace_callback_type_##name(void (*cb)(data_proto))	\
208	{								\
209	}								\
210	static inline bool						\
211	trace_##name##_enabled(void)					\
212	{								\
213		return static_key_false(&__tracepoint_##name.key);	\
214	}
215
216/*
217 * We have no guarantee that gcc and the linker won't up-align the tracepoint
218 * structures, so we create an array of pointers that will be used for iteration
219 * on the tracepoints.
220 */
221#define DEFINE_TRACE_FN(name, reg, unreg)				 \
222	static const char __tpstrtab_##name[]				 \
223	__attribute__((section("__tracepoints_strings"))) = #name;	 \
224	struct tracepoint __tracepoint_##name				 \
225	__attribute__((section("__tracepoints"))) =			 \
226		{ __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
227	static struct tracepoint * const __tracepoint_ptr_##name __used	 \
228	__attribute__((section("__tracepoints_ptrs"))) =		 \
229		&__tracepoint_##name;
230
231#define DEFINE_TRACE(name)						\
232	DEFINE_TRACE_FN(name, NULL, NULL);
233
234#define EXPORT_TRACEPOINT_SYMBOL_GPL(name)				\
235	EXPORT_SYMBOL_GPL(__tracepoint_##name)
236#define EXPORT_TRACEPOINT_SYMBOL(name)					\
237	EXPORT_SYMBOL(__tracepoint_##name)
238
239#else /* !CONFIG_TRACEPOINTS */
240#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
241	static inline void trace_##name(proto)				\
242	{ }								\
243	static inline void trace_##name##_rcuidle(proto)		\
244	{ }								\
245	static inline int						\
246	register_trace_##name(void (*probe)(data_proto),		\
247			      void *data)				\
248	{								\
249		return -ENOSYS;						\
250	}								\
251	static inline int						\
252	unregister_trace_##name(void (*probe)(data_proto),		\
253				void *data)				\
254	{								\
255		return -ENOSYS;						\
256	}								\
257	static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
258	{								\
259	}								\
260	static inline bool						\
261	trace_##name##_enabled(void)					\
262	{								\
263		return false;						\
264	}
265
266#define DEFINE_TRACE_FN(name, reg, unreg)
267#define DEFINE_TRACE(name)
268#define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
269#define EXPORT_TRACEPOINT_SYMBOL(name)
270
271#endif /* CONFIG_TRACEPOINTS */
272
273#ifdef CONFIG_TRACING
274/**
275 * tracepoint_string - register constant persistent string to trace system
276 * @str - a constant persistent string that will be referenced in tracepoints
277 *
278 * If constant strings are being used in tracepoints, it is faster and
279 * more efficient to just save the pointer to the string and reference
280 * that with a printf "%s" instead of saving the string in the ring buffer
281 * and wasting space and time.
282 *
283 * The problem with the above approach is that userspace tools that read
284 * the binary output of the trace buffers do not have access to the string.
285 * Instead they just show the address of the string which is not very
286 * useful to users.
287 *
288 * With tracepoint_string(), the string will be registered to the tracing
289 * system and exported to userspace via the debugfs/tracing/printk_formats
290 * file that maps the string address to the string text. This way userspace
291 * tools that read the binary buffers have a way to map the pointers to
292 * the ASCII strings they represent.
293 *
294 * The @str used must be a constant string and persistent as it would not
295 * make sense to show a string that no longer exists. But it is still fine
296 * to be used with modules, because when modules are unloaded, if they
297 * had tracepoints, the ring buffers are cleared too. As long as the string
298 * does not change during the life of the module, it is fine to use
299 * tracepoint_string() within a module.
300 */
301#define tracepoint_string(str)						\
302	({								\
303		static const char *___tp_str __tracepoint_string = str; \
304		___tp_str;						\
305	})
306#define __tracepoint_string	__attribute__((section("__tracepoint_str")))
307#else
308/*
309 * tracepoint_string() is used to save the string address for userspace
310 * tracing tools. When tracing isn't configured, there's no need to save
311 * anything.
312 */
313# define tracepoint_string(str) str
314# define __tracepoint_string
315#endif
316
317/*
318 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
319 * (void). "void" is a special value in a function prototype and can
320 * not be combined with other arguments. Since the DECLARE_TRACE()
321 * macro adds a data element at the beginning of the prototype,
322 * we need a way to differentiate "(void *data, proto)" from
323 * "(void *data, void)". The second prototype is invalid.
324 *
325 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
326 * and "void *__data" as the callback prototype.
327 *
328 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
329 * "void *__data, proto" as the callback prototype.
330 */
331#define DECLARE_TRACE_NOARGS(name)					\
332	__DECLARE_TRACE(name, void, ,					\
333			cpu_online(raw_smp_processor_id()),		\
334			void *__data, __data)
335
336#define DECLARE_TRACE(name, proto, args)				\
337	__DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),		\
338			cpu_online(raw_smp_processor_id()),		\
339			PARAMS(void *__data, proto),			\
340			PARAMS(__data, args))
341
342#define DECLARE_TRACE_CONDITION(name, proto, args, cond)		\
343	__DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),		\
344			cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
345			PARAMS(void *__data, proto),			\
346			PARAMS(__data, args))
347
348#define TRACE_EVENT_FLAGS(event, flag)
349
350#define TRACE_EVENT_PERF_PERM(event, expr...)
351
352#endif /* DECLARE_TRACE */
353
354#ifndef TRACE_EVENT
355/*
356 * For use with the TRACE_EVENT macro:
357 *
358 * We define a tracepoint, its arguments, its printk format
359 * and its 'fast binary record' layout.
360 *
361 * Firstly, name your tracepoint via TRACE_EVENT(name : the
362 * 'subsystem_event' notation is fine.
363 *
364 * Think about this whole construct as the
365 * 'trace_sched_switch() function' from now on.
366 *
367 *
368 *  TRACE_EVENT(sched_switch,
369 *
370 *	*
371 *	* A function has a regular function arguments
372 *	* prototype, declare it via TP_PROTO():
373 *	*
374 *
375 *	TP_PROTO(struct rq *rq, struct task_struct *prev,
376 *		 struct task_struct *next),
377 *
378 *	*
379 *	* Define the call signature of the 'function'.
380 *	* (Design sidenote: we use this instead of a
381 *	*  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
382 *	*
383 *
384 *	TP_ARGS(rq, prev, next),
385 *
386 *	*
387 *	* Fast binary tracing: define the trace record via
388 *	* TP_STRUCT__entry(). You can think about it like a
389 *	* regular C structure local variable definition.
390 *	*
391 *	* This is how the trace record is structured and will
392 *	* be saved into the ring buffer. These are the fields
393 *	* that will be exposed to user-space in
394 *	* /sys/kernel/debug/tracing/events/<*>/format.
395 *	*
396 *	* The declared 'local variable' is called '__entry'
397 *	*
398 *	* __field(pid_t, prev_prid) is equivalent to a standard declariton:
399 *	*
400 *	*	pid_t	prev_pid;
401 *	*
402 *	* __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
403 *	*
404 *	*	char	prev_comm[TASK_COMM_LEN];
405 *	*
406 *
407 *	TP_STRUCT__entry(
408 *		__array(	char,	prev_comm,	TASK_COMM_LEN	)
409 *		__field(	pid_t,	prev_pid			)
410 *		__field(	int,	prev_prio			)
411 *		__array(	char,	next_comm,	TASK_COMM_LEN	)
412 *		__field(	pid_t,	next_pid			)
413 *		__field(	int,	next_prio			)
414 *	),
415 *
416 *	*
417 *	* Assign the entry into the trace record, by embedding
418 *	* a full C statement block into TP_fast_assign(). You
419 *	* can refer to the trace record as '__entry' -
420 *	* otherwise you can put arbitrary C code in here.
421 *	*
422 *	* Note: this C code will execute every time a trace event
423 *	* happens, on an active tracepoint.
424 *	*
425 *
426 *	TP_fast_assign(
427 *		memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
428 *		__entry->prev_pid	= prev->pid;
429 *		__entry->prev_prio	= prev->prio;
430 *		memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
431 *		__entry->next_pid	= next->pid;
432 *		__entry->next_prio	= next->prio;
433 *	),
434 *
435 *	*
436 *	* Formatted output of a trace record via TP_printk().
437 *	* This is how the tracepoint will appear under ftrace
438 *	* plugins that make use of this tracepoint.
439 *	*
440 *	* (raw-binary tracing wont actually perform this step.)
441 *	*
442 *
443 *	TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
444 *		__entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
445 *		__entry->next_comm, __entry->next_pid, __entry->next_prio),
446 *
447 * );
448 *
449 * This macro construct is thus used for the regular printk format
450 * tracing setup, it is used to construct a function pointer based
451 * tracepoint callback (this is used by programmatic plugins and
452 * can also by used by generic instrumentation like SystemTap), and
453 * it is also used to expose a structured trace record in
454 * /sys/kernel/debug/tracing/events/.
455 *
456 * A set of (un)registration functions can be passed to the variant
457 * TRACE_EVENT_FN to perform any (un)registration work.
458 */
459
460#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
461#define DEFINE_EVENT(template, name, proto, args)		\
462	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
463#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
464	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
465#define DEFINE_EVENT_PRINT(template, name, proto, args, print)	\
466	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
467#define DEFINE_EVENT_CONDITION(template, name, proto,		\
468			       args, cond)			\
469	DECLARE_TRACE_CONDITION(name, PARAMS(proto),		\
470				PARAMS(args), PARAMS(cond))
471
472#define TRACE_EVENT(name, proto, args, struct, assign, print)	\
473	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
474#define TRACE_EVENT_FN(name, proto, args, struct,		\
475		assign, print, reg, unreg)			\
476	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
477#define TRACE_EVENT_CONDITION(name, proto, args, cond,		\
478			      struct, assign, print)		\
479	DECLARE_TRACE_CONDITION(name, PARAMS(proto),		\
480				PARAMS(args), PARAMS(cond))
481
482#define TRACE_EVENT_FLAGS(event, flag)
483
484#define TRACE_EVENT_PERF_PERM(event, expr...)
485
486#endif /* ifdef TRACE_EVENT (see note above) */
487